data for physical capital in 1960, they disappear from the sample. To be sure that our
results do not hinge on a reduced number of countries, we replicate the estimates in
the shorter 1970–1990 period. This increases the sample size to 81 countries. The estimates
that use the Benhabib and Spiegel or Pritchett formulation for human capital
(regressions 6 to 8) display non-significant and negative coefficients. In contrast, the
estimates based on the Mincerian approach are strongly significant. The coefficient
on schooling is now estimated at 12.3% when the initial level of physical capital is
omitted (regression 9) and at 9% when it is included (regression 10). Summing up,
these regressions show that Benhabib and Spiegel and Pritchett failed to find a significant
effect from changes in schooling on growth because their data was too noisy
and becau